2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
10 #include "string-list.h"
15 #include "run-command.h"
18 #include "tree-walk.h"
20 #include "pack-revindex.h"
21 #include "sha1-lookup.h"
22 #include "bulk-checkin.h"
23 #include "streaming.h"
26 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
27 #define O_NOATIME 01000000
33 #define SZ_FMT PRIuMAX
34 static inline uintmax_t sz_fmt(size_t s
) { return s
; }
36 const unsigned char null_sha1
[20];
39 * This is meant to hold a *small* number of objects that you would
40 * want read_sha1_file() to be able to return, but yet you do not want
41 * to write them into the object store (e.g. a browse-only
44 static struct cached_object
{
45 unsigned char sha1
[20];
46 enum object_type type
;
50 static int cached_object_nr
, cached_object_alloc
;
52 static struct cached_object empty_tree
= {
53 EMPTY_TREE_SHA1_BIN_LITERAL
,
59 static struct packed_git
*last_found_pack
;
61 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
64 struct cached_object
*co
= cached_objects
;
66 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
67 if (!hashcmp(co
->sha1
, sha1
))
70 if (!hashcmp(sha1
, empty_tree
.sha1
))
75 int mkdir_in_gitdir(const char *path
)
77 if (mkdir(path
, 0777)) {
78 int saved_errno
= errno
;
80 struct strbuf sb
= STRBUF_INIT
;
85 * Are we looking at a path in a symlinked worktree
86 * whose original repository does not yet have it?
87 * e.g. .git/rr-cache pointing at its original
88 * repository in which the user hasn't performed any
89 * conflict resolution yet?
91 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
92 strbuf_readlink(&sb
, path
, st
.st_size
) ||
93 !is_absolute_path(sb
.buf
) ||
94 mkdir(sb
.buf
, 0777)) {
101 return adjust_shared_perm(path
);
104 int safe_create_leading_directories(char *path
)
106 char *pos
= path
+ offset_1st_component(path
);
110 pos
= strchr(pos
, '/');
113 while (*++pos
== '/')
118 if (!stat(path
, &st
)) {
120 if (!S_ISDIR(st
.st_mode
)) {
125 else if (mkdir(path
, 0777)) {
129 else if (adjust_shared_perm(path
)) {
138 int safe_create_leading_directories_const(const char *path
)
140 /* path points to cache entries, so xstrdup before messing with it */
141 char *buf
= xstrdup(path
);
142 int result
= safe_create_leading_directories(buf
);
147 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
150 for (i
= 0; i
< 20; i
++) {
151 static char hex
[] = "0123456789abcdef";
152 unsigned int val
= sha1
[i
];
153 char *pos
= pathbuf
+ i
*2 + (i
> 0);
154 *pos
++ = hex
[val
>> 4];
155 *pos
= hex
[val
& 0xf];
160 * NOTE! This returns a statically allocated buffer, so you have to be
161 * careful about using it. Do an "xstrdup()" if you need to save the
164 * Also note that this returns the location for creating. Reading
165 * SHA1 file can happen from any alternate directory listed in the
166 * DB_ENVIRONMENT environment variable if it is not found in
167 * the primary object database.
169 char *sha1_file_name(const unsigned char *sha1
)
171 static char buf
[PATH_MAX
];
175 objdir
= get_object_directory();
176 len
= strlen(objdir
);
178 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
179 if (len
+ 43 > PATH_MAX
)
180 die("insanely long object directory %s", objdir
);
181 memcpy(buf
, objdir
, len
);
185 fill_sha1_path(buf
+ len
+ 1, sha1
);
189 static char *sha1_get_pack_name(const unsigned char *sha1
,
190 char **name
, char **base
, const char *which
)
192 static const char hex
[] = "0123456789abcdef";
197 const char *sha1_file_directory
= get_object_directory();
198 int len
= strlen(sha1_file_directory
);
199 *base
= xmalloc(len
+ 60);
200 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
201 sha1_file_directory
, which
);
202 *name
= *base
+ len
+ 11;
207 for (i
= 0; i
< 20; i
++) {
208 unsigned int val
= *sha1
++;
209 *buf
++ = hex
[val
>> 4];
210 *buf
++ = hex
[val
& 0xf];
216 char *sha1_pack_name(const unsigned char *sha1
)
218 static char *name
, *base
;
220 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
223 char *sha1_pack_index_name(const unsigned char *sha1
)
225 static char *name
, *base
;
227 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
230 struct alternate_object_database
*alt_odb_list
;
231 static struct alternate_object_database
**alt_odb_tail
;
233 static int git_open_noatime(const char *name
);
236 * Prepare alternate object database registry.
238 * The variable alt_odb_list points at the list of struct
239 * alternate_object_database. The elements on this list come from
240 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
241 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
242 * whose contents is similar to that environment variable but can be
243 * LF separated. Its base points at a statically allocated buffer that
244 * contains "/the/directory/corresponding/to/.git/objects/...", while
245 * its name points just after the slash at the end of ".git/objects/"
246 * in the example above, and has enough space to hold 40-byte hex
247 * SHA1, an extra slash for the first level indirection, and the
250 static int link_alt_odb_entry(const char *entry
, const char *relative_base
, int depth
)
252 const char *objdir
= get_object_directory();
253 struct alternate_object_database
*ent
;
254 struct alternate_object_database
*alt
;
256 struct strbuf pathbuf
= STRBUF_INIT
;
258 if (!is_absolute_path(entry
) && relative_base
) {
259 strbuf_addstr(&pathbuf
, real_path(relative_base
));
260 strbuf_addch(&pathbuf
, '/');
262 strbuf_addstr(&pathbuf
, entry
);
264 normalize_path_copy(pathbuf
.buf
, pathbuf
.buf
);
266 pfxlen
= strlen(pathbuf
.buf
);
269 * The trailing slash after the directory name is given by
270 * this function at the end. Remove duplicates.
272 while (pfxlen
&& pathbuf
.buf
[pfxlen
-1] == '/')
275 entlen
= pfxlen
+ 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
276 ent
= xmalloc(sizeof(*ent
) + entlen
);
277 memcpy(ent
->base
, pathbuf
.buf
, pfxlen
);
278 strbuf_release(&pathbuf
);
280 ent
->name
= ent
->base
+ pfxlen
+ 1;
281 ent
->base
[pfxlen
+ 3] = '/';
282 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
284 /* Detect cases where alternate disappeared */
285 if (!is_directory(ent
->base
)) {
286 error("object directory %s does not exist; "
287 "check .git/objects/info/alternates.",
293 /* Prevent the common mistake of listing the same
294 * thing twice, or object directory itself.
296 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
297 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
302 if (!strcmp(ent
->base
, objdir
)) {
307 /* add the alternate entry */
309 alt_odb_tail
= &(ent
->next
);
312 /* recursively add alternates */
313 read_info_alternates(ent
->base
, depth
+ 1);
315 ent
->base
[pfxlen
] = '/';
320 static void link_alt_odb_entries(const char *alt
, int len
, int sep
,
321 const char *relative_base
, int depth
)
323 struct string_list entries
= STRING_LIST_INIT_NODUP
;
328 error("%s: ignoring alternate object stores, nesting too deep.",
333 alt_copy
= xmemdupz(alt
, len
);
334 string_list_split_in_place(&entries
, alt_copy
, sep
, -1);
335 for (i
= 0; i
< entries
.nr
; i
++) {
336 const char *entry
= entries
.items
[i
].string
;
337 if (entry
[0] == '\0' || entry
[0] == '#')
339 if (!is_absolute_path(entry
) && depth
) {
340 error("%s: ignoring relative alternate object store %s",
341 relative_base
, entry
);
343 link_alt_odb_entry(entry
, relative_base
, depth
);
346 string_list_clear(&entries
, 0);
350 void read_info_alternates(const char * relative_base
, int depth
)
355 const char alt_file_name
[] = "info/alternates";
356 /* Given that relative_base is no longer than PATH_MAX,
357 ensure that "path" has enough space to append "/", the
358 file name, "info/alternates", and a trailing NUL. */
359 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
362 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
363 fd
= git_open_noatime(path
);
366 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
370 mapsz
= xsize_t(st
.st_size
);
371 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
374 link_alt_odb_entries(map
, mapsz
, '\n', relative_base
, depth
);
379 void add_to_alternates_file(const char *reference
)
381 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
382 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
383 char *alt
= mkpath("%s\n", reference
);
384 write_or_die(fd
, alt
, strlen(alt
));
385 if (commit_lock_file(lock
))
386 die("could not close alternates file");
388 link_alt_odb_entries(alt
, strlen(alt
), '\n', NULL
, 0);
391 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
393 struct alternate_object_database
*ent
;
396 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
401 void prepare_alt_odb(void)
408 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
411 alt_odb_tail
= &alt_odb_list
;
412 link_alt_odb_entries(alt
, strlen(alt
), PATH_SEP
, NULL
, 0);
414 read_info_alternates(get_object_directory(), 0);
417 static int has_loose_object_local(const unsigned char *sha1
)
419 char *name
= sha1_file_name(sha1
);
420 return !access(name
, F_OK
);
423 int has_loose_object_nonlocal(const unsigned char *sha1
)
425 struct alternate_object_database
*alt
;
427 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
428 fill_sha1_path(alt
->name
, sha1
);
429 if (!access(alt
->base
, F_OK
))
435 static int has_loose_object(const unsigned char *sha1
)
437 return has_loose_object_local(sha1
) ||
438 has_loose_object_nonlocal(sha1
);
441 static unsigned int pack_used_ctr
;
442 static unsigned int pack_mmap_calls
;
443 static unsigned int peak_pack_open_windows
;
444 static unsigned int pack_open_windows
;
445 static unsigned int pack_open_fds
;
446 static unsigned int pack_max_fds
;
447 static size_t peak_pack_mapped
;
448 static size_t pack_mapped
;
449 struct packed_git
*packed_git
;
451 void pack_report(void)
454 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
455 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
456 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
457 sz_fmt(getpagesize()),
458 sz_fmt(packed_git_window_size
),
459 sz_fmt(packed_git_limit
));
461 "pack_report: pack_used_ctr = %10u\n"
462 "pack_report: pack_mmap_calls = %10u\n"
463 "pack_report: pack_open_windows = %10u / %10u\n"
464 "pack_report: pack_mapped = "
465 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
468 pack_open_windows
, peak_pack_open_windows
,
469 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
472 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
475 struct pack_idx_header
*hdr
;
477 uint32_t version
, nr
, i
, *index
;
478 int fd
= git_open_noatime(path
);
483 if (fstat(fd
, &st
)) {
487 idx_size
= xsize_t(st
.st_size
);
488 if (idx_size
< 4 * 256 + 20 + 20) {
490 return error("index file %s is too small", path
);
492 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
496 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
497 version
= ntohl(hdr
->idx_version
);
498 if (version
< 2 || version
> 2) {
499 munmap(idx_map
, idx_size
);
500 return error("index file %s is version %"PRIu32
501 " and is not supported by this binary"
502 " (try upgrading GIT to a newer version)",
511 index
+= 2; /* skip index header */
512 for (i
= 0; i
< 256; i
++) {
513 uint32_t n
= ntohl(index
[i
]);
515 munmap(idx_map
, idx_size
);
516 return error("non-monotonic index %s", path
);
524 * - 256 index entries 4 bytes each
525 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
526 * - 20-byte SHA1 of the packfile
527 * - 20-byte SHA1 file checksum
529 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
530 munmap(idx_map
, idx_size
);
531 return error("wrong index v1 file size in %s", path
);
533 } else if (version
== 2) {
536 * - 8 bytes of header
537 * - 256 index entries 4 bytes each
538 * - 20-byte sha1 entry * nr
539 * - 4-byte crc entry * nr
540 * - 4-byte offset entry * nr
541 * - 20-byte SHA1 of the packfile
542 * - 20-byte SHA1 file checksum
543 * And after the 4-byte offset table might be a
544 * variable sized table containing 8-byte entries
545 * for offsets larger than 2^31.
547 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
548 unsigned long max_size
= min_size
;
550 max_size
+= (nr
- 1)*8;
551 if (idx_size
< min_size
|| idx_size
> max_size
) {
552 munmap(idx_map
, idx_size
);
553 return error("wrong index v2 file size in %s", path
);
555 if (idx_size
!= min_size
&&
557 * make sure we can deal with large pack offsets.
558 * 31-bit signed offset won't be enough, neither
559 * 32-bit unsigned one will be.
561 (sizeof(off_t
) <= 4)) {
562 munmap(idx_map
, idx_size
);
563 return error("pack too large for current definition of off_t in %s", path
);
567 p
->index_version
= version
;
568 p
->index_data
= idx_map
;
569 p
->index_size
= idx_size
;
574 int open_pack_index(struct packed_git
*p
)
582 idx_name
= xstrdup(p
->pack_name
);
583 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
584 ret
= check_packed_git_idx(idx_name
, p
);
589 static void scan_windows(struct packed_git
*p
,
590 struct packed_git
**lru_p
,
591 struct pack_window
**lru_w
,
592 struct pack_window
**lru_l
)
594 struct pack_window
*w
, *w_l
;
596 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
598 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
608 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
610 struct packed_git
*p
, *lru_p
= NULL
;
611 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
614 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
615 for (p
= packed_git
; p
; p
= p
->next
)
616 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
618 munmap(lru_w
->base
, lru_w
->len
);
619 pack_mapped
-= lru_w
->len
;
621 lru_l
->next
= lru_w
->next
;
623 lru_p
->windows
= lru_w
->next
;
624 if (!lru_p
->windows
&& lru_p
->pack_fd
!= -1
625 && lru_p
->pack_fd
!= keep_fd
) {
626 close(lru_p
->pack_fd
);
638 void release_pack_memory(size_t need
, int fd
)
640 size_t cur
= pack_mapped
;
641 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
645 void *xmmap(void *start
, size_t length
,
646 int prot
, int flags
, int fd
, off_t offset
)
648 void *ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
649 if (ret
== MAP_FAILED
) {
652 release_pack_memory(length
, fd
);
653 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
654 if (ret
== MAP_FAILED
)
655 die_errno("Out of memory? mmap failed");
660 void close_pack_windows(struct packed_git
*p
)
663 struct pack_window
*w
= p
->windows
;
666 die("pack '%s' still has open windows to it",
668 munmap(w
->base
, w
->len
);
669 pack_mapped
-= w
->len
;
671 p
->windows
= w
->next
;
677 * The LRU pack is the one with the oldest MRU window, preferring packs
678 * with no used windows, or the oldest mtime if it has no windows allocated.
680 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
682 struct pack_window
*w
, *this_mru_w
;
683 int has_windows_inuse
= 0;
686 * Reject this pack if it has windows and the previously selected
687 * one does not. If this pack does not have windows, reject
688 * it if the pack file is newer than the previously selected one.
690 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
693 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
695 * Reject this pack if any of its windows are in use,
696 * but the previously selected pack did not have any
697 * inuse windows. Otherwise, record that this pack
698 * has windows in use.
701 if (*accept_windows_inuse
)
702 has_windows_inuse
= 1;
707 if (w
->last_used
> this_mru_w
->last_used
)
711 * Reject this pack if it has windows that have been
712 * used more recently than the previously selected pack.
713 * If the previously selected pack had windows inuse and
714 * we have not encountered a window in this pack that is
715 * inuse, skip this check since we prefer a pack with no
716 * inuse windows to one that has inuse windows.
718 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
719 this_mru_w
->last_used
> (*mru_w
)->last_used
)
728 *accept_windows_inuse
= has_windows_inuse
;
731 static int close_one_pack(void)
733 struct packed_git
*p
, *lru_p
= NULL
;
734 struct pack_window
*mru_w
= NULL
;
735 int accept_windows_inuse
= 1;
737 for (p
= packed_git
; p
; p
= p
->next
) {
738 if (p
->pack_fd
== -1)
740 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
744 close(lru_p
->pack_fd
);
753 void unuse_pack(struct pack_window
**w_cursor
)
755 struct pack_window
*w
= *w_cursor
;
762 void close_pack_index(struct packed_git
*p
)
765 munmap((void *)p
->index_data
, p
->index_size
);
766 p
->index_data
= NULL
;
771 * This is used by git-repack in case a newly created pack happens to
772 * contain the same set of objects as an existing one. In that case
773 * the resulting file might be different even if its name would be the
774 * same. It is best to close any reference to the old pack before it is
775 * replaced on disk. Of course no index pointers nor windows for given pack
776 * must subsist at this point. If ever objects from this pack are requested
777 * again, the new version of the pack will be reinitialized through
778 * reprepare_packed_git().
780 void free_pack_by_name(const char *pack_name
)
782 struct packed_git
*p
, **pp
= &packed_git
;
786 if (strcmp(pack_name
, p
->pack_name
) == 0) {
787 clear_delta_base_cache();
788 close_pack_windows(p
);
789 if (p
->pack_fd
!= -1) {
794 free(p
->bad_object_sha1
);
796 if (last_found_pack
== p
)
797 last_found_pack
= NULL
;
805 static unsigned int get_max_fd_limit(void)
810 if (getrlimit(RLIMIT_NOFILE
, &lim
))
811 die_errno("cannot get RLIMIT_NOFILE");
814 #elif defined(_SC_OPEN_MAX)
815 return sysconf(_SC_OPEN_MAX
);
816 #elif defined(OPEN_MAX)
819 return 1; /* see the caller ;-) */
824 * Do not call this directly as this leaks p->pack_fd on error return;
825 * call open_packed_git() instead.
827 static int open_packed_git_1(struct packed_git
*p
)
830 struct pack_header hdr
;
831 unsigned char sha1
[20];
832 unsigned char *idx_sha1
;
835 if (!p
->index_data
&& open_pack_index(p
))
836 return error("packfile %s index unavailable", p
->pack_name
);
839 unsigned int max_fds
= get_max_fd_limit();
841 /* Save 3 for stdin/stdout/stderr, 22 for work */
843 pack_max_fds
= max_fds
- 25;
848 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
851 p
->pack_fd
= git_open_noatime(p
->pack_name
);
852 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
856 /* If we created the struct before we had the pack we lack size. */
858 if (!S_ISREG(st
.st_mode
))
859 return error("packfile %s not a regular file", p
->pack_name
);
860 p
->pack_size
= st
.st_size
;
861 } else if (p
->pack_size
!= st
.st_size
)
862 return error("packfile %s size changed", p
->pack_name
);
864 /* We leave these file descriptors open with sliding mmap;
865 * there is no point keeping them open across exec(), though.
867 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
869 return error("cannot determine file descriptor flags");
870 fd_flag
|= FD_CLOEXEC
;
871 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
872 return error("cannot set FD_CLOEXEC");
874 /* Verify we recognize this pack file format. */
875 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
876 return error("file %s is far too short to be a packfile", p
->pack_name
);
877 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
878 return error("file %s is not a GIT packfile", p
->pack_name
);
879 if (!pack_version_ok(hdr
.hdr_version
))
880 return error("packfile %s is version %"PRIu32
" and not"
881 " supported (try upgrading GIT to a newer version)",
882 p
->pack_name
, ntohl(hdr
.hdr_version
));
884 /* Verify the pack matches its index. */
885 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
886 return error("packfile %s claims to have %"PRIu32
" objects"
887 " while index indicates %"PRIu32
" objects",
888 p
->pack_name
, ntohl(hdr
.hdr_entries
),
890 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
891 return error("end of packfile %s is unavailable", p
->pack_name
);
892 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
893 return error("packfile %s signature is unavailable", p
->pack_name
);
894 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
895 if (hashcmp(sha1
, idx_sha1
))
896 return error("packfile %s does not match index", p
->pack_name
);
900 static int open_packed_git(struct packed_git
*p
)
902 if (!open_packed_git_1(p
))
904 if (p
->pack_fd
!= -1) {
912 static int in_window(struct pack_window
*win
, off_t offset
)
914 /* We must promise at least 20 bytes (one hash) after the
915 * offset is available from this window, otherwise the offset
916 * is not actually in this window and a different window (which
917 * has that one hash excess) must be used. This is to support
918 * the object header and delta base parsing routines below.
920 off_t win_off
= win
->offset
;
921 return win_off
<= offset
922 && (offset
+ 20) <= (win_off
+ win
->len
);
925 unsigned char *use_pack(struct packed_git
*p
,
926 struct pack_window
**w_cursor
,
930 struct pack_window
*win
= *w_cursor
;
932 /* Since packfiles end in a hash of their content and it's
933 * pointless to ask for an offset into the middle of that
934 * hash, and the in_window function above wouldn't match
935 * don't allow an offset too close to the end of the file.
937 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
938 die("packfile %s cannot be accessed", p
->pack_name
);
939 if (offset
> (p
->pack_size
- 20))
940 die("offset beyond end of packfile (truncated pack?)");
942 if (!win
|| !in_window(win
, offset
)) {
945 for (win
= p
->windows
; win
; win
= win
->next
) {
946 if (in_window(win
, offset
))
950 size_t window_align
= packed_git_window_size
/ 2;
953 if (p
->pack_fd
== -1 && open_packed_git(p
))
954 die("packfile %s cannot be accessed", p
->pack_name
);
956 win
= xcalloc(1, sizeof(*win
));
957 win
->offset
= (offset
/ window_align
) * window_align
;
958 len
= p
->pack_size
- win
->offset
;
959 if (len
> packed_git_window_size
)
960 len
= packed_git_window_size
;
961 win
->len
= (size_t)len
;
962 pack_mapped
+= win
->len
;
963 while (packed_git_limit
< pack_mapped
964 && unuse_one_window(p
, p
->pack_fd
))
966 win
->base
= xmmap(NULL
, win
->len
,
967 PROT_READ
, MAP_PRIVATE
,
968 p
->pack_fd
, win
->offset
);
969 if (win
->base
== MAP_FAILED
)
970 die("packfile %s cannot be mapped: %s",
973 if (!win
->offset
&& win
->len
== p
->pack_size
974 && !p
->do_not_close
) {
981 if (pack_mapped
> peak_pack_mapped
)
982 peak_pack_mapped
= pack_mapped
;
983 if (pack_open_windows
> peak_pack_open_windows
)
984 peak_pack_open_windows
= pack_open_windows
;
985 win
->next
= p
->windows
;
989 if (win
!= *w_cursor
) {
990 win
->last_used
= pack_used_ctr
++;
994 offset
-= win
->offset
;
996 *left
= win
->len
- xsize_t(offset
);
997 return win
->base
+ offset
;
1000 static struct packed_git
*alloc_packed_git(int extra
)
1002 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
1003 memset(p
, 0, sizeof(*p
));
1008 static void try_to_free_pack_memory(size_t size
)
1010 release_pack_memory(size
, -1);
1013 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
1015 static int have_set_try_to_free_routine
;
1017 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
1019 if (!have_set_try_to_free_routine
) {
1020 have_set_try_to_free_routine
= 1;
1021 set_try_to_free_routine(try_to_free_pack_memory
);
1025 * Make sure a corresponding .pack file exists and that
1026 * the index looks sane.
1028 path_len
-= strlen(".idx");
1033 memcpy(p
->pack_name
, path
, path_len
);
1035 strcpy(p
->pack_name
+ path_len
, ".keep");
1036 if (!access(p
->pack_name
, F_OK
))
1039 strcpy(p
->pack_name
+ path_len
, ".pack");
1040 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
1045 /* ok, it looks sane as far as we can check without
1046 * actually mapping the pack file.
1048 p
->pack_size
= st
.st_size
;
1049 p
->pack_local
= local
;
1050 p
->mtime
= st
.st_mtime
;
1051 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
1056 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
1058 const char *path
= sha1_pack_name(sha1
);
1059 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
1061 strcpy(p
->pack_name
, path
);
1062 hashcpy(p
->sha1
, sha1
);
1063 if (check_packed_git_idx(idx_path
, p
)) {
1071 void install_packed_git(struct packed_git
*pack
)
1073 if (pack
->pack_fd
!= -1)
1076 pack
->next
= packed_git
;
1080 static void prepare_packed_git_one(char *objdir
, int local
)
1082 /* Ensure that this buffer is large enough so that we can
1083 append "/pack/" without clobbering the stack even if
1084 strlen(objdir) were PATH_MAX. */
1085 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
1090 sprintf(path
, "%s/pack", objdir
);
1092 dir
= opendir(path
);
1094 if (errno
!= ENOENT
)
1095 error("unable to open object pack directory: %s: %s",
1096 path
, strerror(errno
));
1100 while ((de
= readdir(dir
)) != NULL
) {
1101 int namelen
= strlen(de
->d_name
);
1102 struct packed_git
*p
;
1104 if (!has_extension(de
->d_name
, ".idx"))
1107 if (len
+ namelen
+ 1 > sizeof(path
))
1110 /* Don't reopen a pack we already have. */
1111 strcpy(path
+ len
, de
->d_name
);
1112 for (p
= packed_git
; p
; p
= p
->next
) {
1113 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
1118 /* See if it really is a valid .idx file with corresponding
1119 * .pack file that we can map.
1121 p
= add_packed_git(path
, len
+ namelen
, local
);
1124 install_packed_git(p
);
1129 static int sort_pack(const void *a_
, const void *b_
)
1131 struct packed_git
*a
= *((struct packed_git
**)a_
);
1132 struct packed_git
*b
= *((struct packed_git
**)b_
);
1136 * Local packs tend to contain objects specific to our
1137 * variant of the project than remote ones. In addition,
1138 * remote ones could be on a network mounted filesystem.
1139 * Favor local ones for these reasons.
1141 st
= a
->pack_local
- b
->pack_local
;
1146 * Younger packs tend to contain more recent objects,
1147 * and more recent objects tend to get accessed more
1150 if (a
->mtime
< b
->mtime
)
1152 else if (a
->mtime
== b
->mtime
)
1157 static void rearrange_packed_git(void)
1159 struct packed_git
**ary
, *p
;
1162 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1167 /* prepare an array of packed_git for easier sorting */
1168 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1169 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1172 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1174 /* link them back again */
1175 for (i
= 0; i
< n
- 1; i
++)
1176 ary
[i
]->next
= ary
[i
+ 1];
1177 ary
[n
- 1]->next
= NULL
;
1178 packed_git
= ary
[0];
1183 static int prepare_packed_git_run_once
= 0;
1184 void prepare_packed_git(void)
1186 struct alternate_object_database
*alt
;
1188 if (prepare_packed_git_run_once
)
1190 prepare_packed_git_one(get_object_directory(), 1);
1192 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1194 prepare_packed_git_one(alt
->base
, 0);
1195 alt
->name
[-1] = '/';
1197 rearrange_packed_git();
1198 prepare_packed_git_run_once
= 1;
1201 void reprepare_packed_git(void)
1204 prepare_packed_git_run_once
= 0;
1205 prepare_packed_git();
1208 static void mark_bad_packed_object(struct packed_git
*p
,
1209 const unsigned char *sha1
)
1212 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1213 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1215 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1216 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1217 p
->num_bad_objects
++;
1220 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1222 struct packed_git
*p
;
1225 for (p
= packed_git
; p
; p
= p
->next
)
1226 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1227 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1233 * With an in-core object data in "map", rehash it to make sure the
1234 * object name actually matches "sha1" to detect object corruption.
1235 * With "map" == NULL, try reading the object named with "sha1" using
1236 * the streaming interface and rehash it to do the same.
1238 int check_sha1_signature(const unsigned char *sha1
, void *map
,
1239 unsigned long size
, const char *type
)
1241 unsigned char real_sha1
[20];
1242 enum object_type obj_type
;
1243 struct git_istream
*st
;
1249 hash_sha1_file(map
, size
, type
, real_sha1
);
1250 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1253 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
1257 /* Generate the header */
1258 hdrlen
= sprintf(hdr
, "%s %lu", typename(obj_type
), size
) + 1;
1262 git_SHA1_Update(&c
, hdr
, hdrlen
);
1264 char buf
[1024 * 16];
1265 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1269 git_SHA1_Update(&c
, buf
, readlen
);
1271 git_SHA1_Final(real_sha1
, &c
);
1273 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1276 static int git_open_noatime(const char *name
)
1278 static int sha1_file_open_flag
= O_NOATIME
;
1281 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1285 /* Might the failure be due to O_NOATIME? */
1286 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1287 sha1_file_open_flag
= 0;
1295 static int open_sha1_file(const unsigned char *sha1
)
1298 char *name
= sha1_file_name(sha1
);
1299 struct alternate_object_database
*alt
;
1301 fd
= git_open_noatime(name
);
1307 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1309 fill_sha1_path(name
, sha1
);
1310 fd
= git_open_noatime(alt
->base
);
1317 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1322 fd
= open_sha1_file(sha1
);
1327 if (!fstat(fd
, &st
)) {
1328 *size
= xsize_t(st
.st_size
);
1330 /* mmap() is forbidden on empty files */
1331 error("object file %s is empty", sha1_file_name(sha1
));
1334 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1342 * There used to be a second loose object header format which
1343 * was meant to mimic the in-pack format, allowing for direct
1344 * copy of the object data. This format turned up not to be
1345 * really worth it and we no longer write loose objects in that
1348 static int experimental_loose_object(unsigned char *map
)
1353 * We must determine if the buffer contains the standard
1354 * zlib-deflated stream or the experimental format based
1355 * on the in-pack object format. Compare the header byte
1358 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1359 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1361 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1362 * in standard loose-object format, UNLESS it is a Git-pack
1363 * format object *exactly* 8 bytes in size when inflated.
1365 * However, RFC1950 also specifies that the 1st 16-bit word
1366 * must be divisible by 31 - this checksum tells us our buffer
1367 * is in the standard format, giving a false positive only if
1368 * the 1st word of the Git-pack format object happens to be
1369 * divisible by 31, ie:
1370 * ((byte0 * 256) + byte1) % 31 = 0
1371 * => 0ttt10000www1000 % 31 = 0
1373 * As it happens, this case can only arise for www=3 & ttt=1
1374 * - ie, a Commit object, which would have to be 8 bytes in
1375 * size. As no Commit can be that small, we find that the
1376 * combination of these two criteria (bitmask & checksum)
1377 * can always correctly determine the buffer format.
1379 word
= (map
[0] << 8) + map
[1];
1380 if ((map
[0] & 0x8F) == 0x08 && !(word
% 31))
1386 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1387 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1390 unsigned long size
, c
;
1391 unsigned long used
= 0;
1394 *type
= (c
>> 4) & 7;
1398 if (len
<= used
|| bitsizeof(long) <= shift
) {
1399 error("bad object header");
1404 size
+= (c
& 0x7f) << shift
;
1411 int unpack_sha1_header(git_zstream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1413 unsigned long size
, used
;
1414 static const char valid_loose_object_type
[8] = {
1416 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1417 0, /* "delta" and others are invalid in a loose object */
1419 enum object_type type
;
1421 /* Get the data stream */
1422 memset(stream
, 0, sizeof(*stream
));
1423 stream
->next_in
= map
;
1424 stream
->avail_in
= mapsize
;
1425 stream
->next_out
= buffer
;
1426 stream
->avail_out
= bufsiz
;
1428 if (experimental_loose_object(map
)) {
1430 * The old experimental format we no longer produce;
1431 * we can still read it.
1433 used
= unpack_object_header_buffer(map
, mapsize
, &type
, &size
);
1434 if (!used
|| !valid_loose_object_type
[type
])
1439 /* Set up the stream for the rest.. */
1440 stream
->next_in
= map
;
1441 stream
->avail_in
= mapsize
;
1442 git_inflate_init(stream
);
1444 /* And generate the fake traditional header */
1445 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1446 typename(type
), size
);
1449 git_inflate_init(stream
);
1450 return git_inflate(stream
, 0);
1453 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1455 int bytes
= strlen(buffer
) + 1;
1456 unsigned char *buf
= xmallocz(size
);
1460 n
= stream
->total_out
- bytes
;
1463 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1465 if (bytes
<= size
) {
1467 * The above condition must be (bytes <= size), not
1468 * (bytes < size). In other words, even though we
1469 * expect no more output and set avail_out to zero,
1470 * the input zlib stream may have bytes that express
1471 * "this concludes the stream", and we *do* want to
1474 * Otherwise we would not be able to test that we
1475 * consumed all the input to reach the expected size;
1476 * we also want to check that zlib tells us that all
1477 * went well with status == Z_STREAM_END at the end.
1479 stream
->next_out
= buf
+ bytes
;
1480 stream
->avail_out
= size
- bytes
;
1481 while (status
== Z_OK
)
1482 status
= git_inflate(stream
, Z_FINISH
);
1484 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1485 git_inflate_end(stream
);
1490 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1491 else if (stream
->avail_in
)
1492 error("garbage at end of loose object '%s'",
1499 * We used to just use "sscanf()", but that's actually way
1500 * too permissive for what we want to check. So do an anal
1501 * object header parse by hand.
1503 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1510 * The type can be at most ten bytes (including the
1511 * terminating '\0' that we add), and is followed by
1520 if (i
>= sizeof(type
))
1526 * The length must follow immediately, and be in canonical
1527 * decimal format (ie "010" is not valid).
1529 size
= *hdr
++ - '0';
1534 unsigned long c
= *hdr
- '0';
1538 size
= size
* 10 + c
;
1544 * The length must be followed by a zero byte
1546 return *hdr
? -1 : type_from_string(type
);
1549 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1555 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1556 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1559 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1562 unsigned long get_size_from_delta(struct packed_git
*p
,
1563 struct pack_window
**w_curs
,
1566 const unsigned char *data
;
1567 unsigned char delta_head
[20], *in
;
1571 memset(&stream
, 0, sizeof(stream
));
1572 stream
.next_out
= delta_head
;
1573 stream
.avail_out
= sizeof(delta_head
);
1575 git_inflate_init(&stream
);
1577 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1578 stream
.next_in
= in
;
1579 st
= git_inflate(&stream
, Z_FINISH
);
1580 curpos
+= stream
.next_in
- in
;
1581 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1582 stream
.total_out
< sizeof(delta_head
));
1583 git_inflate_end(&stream
);
1584 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1585 error("delta data unpack-initial failed");
1589 /* Examine the initial part of the delta to figure out
1594 /* ignore base size */
1595 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1597 /* Read the result size */
1598 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1601 static off_t
get_delta_base(struct packed_git
*p
,
1602 struct pack_window
**w_curs
,
1604 enum object_type type
,
1605 off_t delta_obj_offset
)
1607 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1610 /* use_pack() assured us we have [base_info, base_info + 20)
1611 * as a range that we can look at without walking off the
1612 * end of the mapped window. Its actually the hash size
1613 * that is assured. An OFS_DELTA longer than the hash size
1614 * is stupid, as then a REF_DELTA would be smaller to store.
1616 if (type
== OBJ_OFS_DELTA
) {
1618 unsigned char c
= base_info
[used
++];
1619 base_offset
= c
& 127;
1622 if (!base_offset
|| MSB(base_offset
, 7))
1623 return 0; /* overflow */
1624 c
= base_info
[used
++];
1625 base_offset
= (base_offset
<< 7) + (c
& 127);
1627 base_offset
= delta_obj_offset
- base_offset
;
1628 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1629 return 0; /* out of bound */
1631 } else if (type
== OBJ_REF_DELTA
) {
1632 /* The base entry _must_ be in the same pack */
1633 base_offset
= find_pack_entry_one(base_info
, p
);
1636 die("I am totally screwed");
1640 /* forward declaration for a mutually recursive function */
1641 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1642 unsigned long *sizep
, int *rtype
);
1644 static int packed_delta_info(struct packed_git
*p
,
1645 struct pack_window
**w_curs
,
1647 enum object_type type
,
1649 unsigned long *sizep
)
1653 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1656 type
= packed_object_info(p
, base_offset
, NULL
, NULL
);
1657 if (type
<= OBJ_NONE
) {
1658 struct revindex_entry
*revidx
;
1659 const unsigned char *base_sha1
;
1660 revidx
= find_pack_revindex(p
, base_offset
);
1663 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1664 mark_bad_packed_object(p
, base_sha1
);
1665 type
= sha1_object_info(base_sha1
, NULL
);
1666 if (type
<= OBJ_NONE
)
1670 /* We choose to only get the type of the base object and
1671 * ignore potentially corrupt pack file that expects the delta
1672 * based on a base with a wrong size. This saves tons of
1676 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1684 int unpack_object_header(struct packed_git
*p
,
1685 struct pack_window
**w_curs
,
1687 unsigned long *sizep
)
1689 unsigned char *base
;
1692 enum object_type type
;
1694 /* use_pack() assures us we have [base, base + 20) available
1695 * as a range that we can look at. (Its actually the hash
1696 * size that is assured.) With our object header encoding
1697 * the maximum deflated object size is 2^137, which is just
1698 * insane, so we know won't exceed what we have been given.
1700 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1701 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1710 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1711 unsigned long *sizep
, int *rtype
)
1713 struct pack_window
*w_curs
= NULL
;
1715 off_t curpos
= obj_offset
;
1716 enum object_type type
;
1718 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1720 *rtype
= type
; /* representation type */
1725 type
= packed_delta_info(p
, &w_curs
, curpos
,
1726 type
, obj_offset
, sizep
);
1736 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1737 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1740 unuse_pack(&w_curs
);
1744 static void *unpack_compressed_entry(struct packed_git
*p
,
1745 struct pack_window
**w_curs
,
1751 unsigned char *buffer
, *in
;
1753 buffer
= xmallocz(size
);
1754 memset(&stream
, 0, sizeof(stream
));
1755 stream
.next_out
= buffer
;
1756 stream
.avail_out
= size
+ 1;
1758 git_inflate_init(&stream
);
1760 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1761 stream
.next_in
= in
;
1762 st
= git_inflate(&stream
, Z_FINISH
);
1763 if (!stream
.avail_out
)
1764 break; /* the payload is larger than it should be */
1765 curpos
+= stream
.next_in
- in
;
1766 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1767 git_inflate_end(&stream
);
1768 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1776 #define MAX_DELTA_CACHE (256)
1778 static size_t delta_base_cached
;
1780 static struct delta_base_cache_lru_list
{
1781 struct delta_base_cache_lru_list
*prev
;
1782 struct delta_base_cache_lru_list
*next
;
1783 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1785 static struct delta_base_cache_entry
{
1786 struct delta_base_cache_lru_list lru
;
1788 struct packed_git
*p
;
1791 enum object_type type
;
1792 } delta_base_cache
[MAX_DELTA_CACHE
];
1794 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1798 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1799 hash
+= (hash
>> 8) + (hash
>> 16);
1800 return hash
% MAX_DELTA_CACHE
;
1803 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
1805 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1806 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1807 return (ent
->data
&& ent
->p
== p
&& ent
->base_offset
== base_offset
);
1810 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1811 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1814 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1815 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1818 if (!ret
|| ent
->p
!= p
|| ent
->base_offset
!= base_offset
)
1819 return unpack_entry(p
, base_offset
, type
, base_size
);
1823 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1824 ent
->lru
.prev
->next
= ent
->lru
.next
;
1825 delta_base_cached
-= ent
->size
;
1827 ret
= xmemdupz(ent
->data
, ent
->size
);
1830 *base_size
= ent
->size
;
1834 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1839 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1840 ent
->lru
.prev
->next
= ent
->lru
.next
;
1841 delta_base_cached
-= ent
->size
;
1845 void clear_delta_base_cache(void)
1848 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
1849 release_delta_base_cache(&delta_base_cache
[p
]);
1852 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1853 void *base
, unsigned long base_size
, enum object_type type
)
1855 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1856 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1857 struct delta_base_cache_lru_list
*lru
;
1859 release_delta_base_cache(ent
);
1860 delta_base_cached
+= base_size
;
1862 for (lru
= delta_base_cache_lru
.next
;
1863 delta_base_cached
> delta_base_cache_limit
1864 && lru
!= &delta_base_cache_lru
;
1866 struct delta_base_cache_entry
*f
= (void *)lru
;
1867 if (f
->type
== OBJ_BLOB
)
1868 release_delta_base_cache(f
);
1870 for (lru
= delta_base_cache_lru
.next
;
1871 delta_base_cached
> delta_base_cache_limit
1872 && lru
!= &delta_base_cache_lru
;
1874 struct delta_base_cache_entry
*f
= (void *)lru
;
1875 release_delta_base_cache(f
);
1879 ent
->base_offset
= base_offset
;
1882 ent
->size
= base_size
;
1883 ent
->lru
.next
= &delta_base_cache_lru
;
1884 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1885 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1886 delta_base_cache_lru
.prev
= &ent
->lru
;
1889 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1890 unsigned long *size
);
1892 static void *unpack_delta_entry(struct packed_git
*p
,
1893 struct pack_window
**w_curs
,
1895 unsigned long delta_size
,
1897 enum object_type
*type
,
1898 unsigned long *sizep
)
1900 void *delta_data
, *result
, *base
;
1901 unsigned long base_size
;
1904 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1906 error("failed to validate delta base reference "
1907 "at offset %"PRIuMAX
" from %s",
1908 (uintmax_t)curpos
, p
->pack_name
);
1912 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1915 * We're probably in deep shit, but let's try to fetch
1916 * the required base anyway from another pack or loose.
1917 * This is costly but should happen only in the presence
1918 * of a corrupted pack, and is better than failing outright.
1920 struct revindex_entry
*revidx
;
1921 const unsigned char *base_sha1
;
1922 revidx
= find_pack_revindex(p
, base_offset
);
1925 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1926 error("failed to read delta base object %s"
1927 " at offset %"PRIuMAX
" from %s",
1928 sha1_to_hex(base_sha1
), (uintmax_t)base_offset
,
1930 mark_bad_packed_object(p
, base_sha1
);
1931 base
= read_object(base_sha1
, type
, &base_size
);
1936 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1938 error("failed to unpack compressed delta "
1939 "at offset %"PRIuMAX
" from %s",
1940 (uintmax_t)curpos
, p
->pack_name
);
1944 result
= patch_delta(base
, base_size
,
1945 delta_data
, delta_size
,
1948 die("failed to apply delta");
1950 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1954 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
1956 static FILE *log_file
;
1959 log_file
= fopen(log_pack_access
, "w");
1961 error("cannot open pack access log '%s' for writing: %s",
1962 log_pack_access
, strerror(errno
));
1963 log_pack_access
= NULL
;
1967 fprintf(log_file
, "%s %"PRIuMAX
"\n",
1968 p
->pack_name
, (uintmax_t)obj_offset
);
1972 int do_check_packed_object_crc
;
1974 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1975 enum object_type
*type
, unsigned long *sizep
)
1977 struct pack_window
*w_curs
= NULL
;
1978 off_t curpos
= obj_offset
;
1981 if (log_pack_access
)
1982 write_pack_access_log(p
, obj_offset
);
1984 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1985 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1986 unsigned long len
= revidx
[1].offset
- obj_offset
;
1987 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1988 const unsigned char *sha1
=
1989 nth_packed_object_sha1(p
, revidx
->nr
);
1990 error("bad packed object CRC for %s",
1992 mark_bad_packed_object(p
, sha1
);
1993 unuse_pack(&w_curs
);
1998 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
2002 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
2003 obj_offset
, type
, sizep
);
2009 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
2013 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2014 *type
, (uintmax_t)obj_offset
, p
->pack_name
);
2016 unuse_pack(&w_curs
);
2020 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
2023 const unsigned char *index
= p
->index_data
;
2025 if (open_pack_index(p
))
2027 index
= p
->index_data
;
2029 if (n
>= p
->num_objects
)
2032 if (p
->index_version
== 1) {
2033 return index
+ 24 * n
+ 4;
2036 return index
+ 20 * n
;
2040 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
2042 const unsigned char *index
= p
->index_data
;
2044 if (p
->index_version
== 1) {
2045 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
2048 index
+= 8 + p
->num_objects
* (20 + 4);
2049 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
2050 if (!(off
& 0x80000000))
2052 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
2053 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
2054 ntohl(*((uint32_t *)(index
+ 4)));
2058 off_t
find_pack_entry_one(const unsigned char *sha1
,
2059 struct packed_git
*p
)
2061 const uint32_t *level1_ofs
= p
->index_data
;
2062 const unsigned char *index
= p
->index_data
;
2063 unsigned hi
, lo
, stride
;
2064 static int use_lookup
= -1;
2065 static int debug_lookup
= -1;
2067 if (debug_lookup
< 0)
2068 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2071 if (open_pack_index(p
))
2073 level1_ofs
= p
->index_data
;
2074 index
= p
->index_data
;
2076 if (p
->index_version
> 1) {
2081 hi
= ntohl(level1_ofs
[*sha1
]);
2082 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2083 if (p
->index_version
> 1) {
2091 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2092 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2095 use_lookup
= !!getenv("GIT_USE_LOOKUP");
2097 int pos
= sha1_entry_pos(index
, stride
, 0,
2098 lo
, hi
, p
->num_objects
, sha1
);
2101 return nth_packed_object_offset(p
, pos
);
2105 unsigned mi
= (lo
+ hi
) / 2;
2106 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2109 printf("lo %u hi %u rg %u mi %u\n",
2110 lo
, hi
, hi
- lo
, mi
);
2112 return nth_packed_object_offset(p
, mi
);
2121 int is_pack_valid(struct packed_git
*p
)
2123 /* An already open pack is known to be valid. */
2124 if (p
->pack_fd
!= -1)
2127 /* If the pack has one window completely covering the
2128 * file size, the pack is known to be valid even if
2129 * the descriptor is not currently open.
2132 struct pack_window
*w
= p
->windows
;
2134 if (!w
->offset
&& w
->len
== p
->pack_size
)
2138 /* Force the pack to open to prove its valid. */
2139 return !open_packed_git(p
);
2142 static int fill_pack_entry(const unsigned char *sha1
,
2143 struct pack_entry
*e
,
2144 struct packed_git
*p
)
2148 if (p
->num_bad_objects
) {
2150 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2151 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2155 offset
= find_pack_entry_one(sha1
, p
);
2160 * We are about to tell the caller where they can locate the
2161 * requested object. We better make sure the packfile is
2162 * still here and can be accessed before supplying that
2163 * answer, as it may have been deleted since the index was
2166 if (!is_pack_valid(p
)) {
2167 warning("packfile %s cannot be accessed", p
->pack_name
);
2172 hashcpy(e
->sha1
, sha1
);
2176 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2178 struct packed_git
*p
;
2180 prepare_packed_git();
2184 if (last_found_pack
&& fill_pack_entry(sha1
, e
, last_found_pack
))
2187 for (p
= packed_git
; p
; p
= p
->next
) {
2188 if (p
== last_found_pack
|| !fill_pack_entry(sha1
, e
, p
))
2191 last_found_pack
= p
;
2197 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2198 struct packed_git
*packs
)
2200 struct packed_git
*p
;
2202 for (p
= packs
; p
; p
= p
->next
) {
2203 if (find_pack_entry_one(sha1
, p
))
2210 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2213 unsigned long mapsize
, size
;
2218 map
= map_sha1_file(sha1
, &mapsize
);
2220 return error("unable to find %s", sha1_to_hex(sha1
));
2221 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2222 status
= error("unable to unpack %s header",
2224 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
2225 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2228 git_inflate_end(&stream
);
2229 munmap(map
, mapsize
);
2233 /* returns enum object_type or negative */
2234 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
)
2236 struct cached_object
*co
;
2237 struct pack_entry e
;
2240 co
= find_cached_object(sha1
);
2243 *(oi
->sizep
) = co
->size
;
2244 oi
->whence
= OI_CACHED
;
2248 if (!find_pack_entry(sha1
, &e
)) {
2249 /* Most likely it's a loose object. */
2250 status
= sha1_loose_object_info(sha1
, oi
->sizep
);
2252 oi
->whence
= OI_LOOSE
;
2256 /* Not a loose object; someone else may have just packed it. */
2257 reprepare_packed_git();
2258 if (!find_pack_entry(sha1
, &e
))
2262 status
= packed_object_info(e
.p
, e
.offset
, oi
->sizep
, &rtype
);
2264 mark_bad_packed_object(e
.p
, sha1
);
2265 status
= sha1_object_info_extended(sha1
, oi
);
2266 } else if (in_delta_base_cache(e
.p
, e
.offset
)) {
2267 oi
->whence
= OI_DBCACHED
;
2269 oi
->whence
= OI_PACKED
;
2270 oi
->u
.packed
.offset
= e
.offset
;
2271 oi
->u
.packed
.pack
= e
.p
;
2272 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
2273 rtype
== OBJ_OFS_DELTA
);
2279 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2281 struct object_info oi
;
2284 return sha1_object_info_extended(sha1
, &oi
);
2287 static void *read_packed_sha1(const unsigned char *sha1
,
2288 enum object_type
*type
, unsigned long *size
)
2290 struct pack_entry e
;
2293 if (!find_pack_entry(sha1
, &e
))
2295 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2298 * We're probably in deep shit, but let's try to fetch
2299 * the required object anyway from another pack or loose.
2300 * This should happen only in the presence of a corrupted
2301 * pack, and is better than failing outright.
2303 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2304 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2305 mark_bad_packed_object(e
.p
, sha1
);
2306 data
= read_object(sha1
, type
, size
);
2311 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2312 unsigned char *sha1
)
2314 struct cached_object
*co
;
2316 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2317 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2319 if (cached_object_alloc
<= cached_object_nr
) {
2320 cached_object_alloc
= alloc_nr(cached_object_alloc
);
2321 cached_objects
= xrealloc(cached_objects
,
2322 sizeof(*cached_objects
) *
2323 cached_object_alloc
);
2325 co
= &cached_objects
[cached_object_nr
++];
2328 co
->buf
= xmalloc(len
);
2329 memcpy(co
->buf
, buf
, len
);
2330 hashcpy(co
->sha1
, sha1
);
2334 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2335 unsigned long *size
)
2337 unsigned long mapsize
;
2339 struct cached_object
*co
;
2341 co
= find_cached_object(sha1
);
2345 return xmemdupz(co
->buf
, co
->size
);
2348 buf
= read_packed_sha1(sha1
, type
, size
);
2351 map
= map_sha1_file(sha1
, &mapsize
);
2353 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2354 munmap(map
, mapsize
);
2357 reprepare_packed_git();
2358 return read_packed_sha1(sha1
, type
, size
);
2362 * This function dies on corrupt objects; the callers who want to
2363 * deal with them should arrange to call read_object() and give error
2364 * messages themselves.
2366 void *read_sha1_file_extended(const unsigned char *sha1
,
2367 enum object_type
*type
,
2368 unsigned long *size
,
2373 const struct packed_git
*p
;
2374 const unsigned char *repl
= (flag
& READ_SHA1_FILE_REPLACE
)
2375 ? lookup_replace_object(sha1
) : sha1
;
2378 data
= read_object(repl
, type
, size
);
2382 if (errno
&& errno
!= ENOENT
)
2383 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2385 /* die if we replaced an object with one that does not exist */
2387 die("replacement %s not found for %s",
2388 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2390 if (has_loose_object(repl
)) {
2391 path
= sha1_file_name(sha1
);
2392 die("loose object %s (stored in %s) is corrupt",
2393 sha1_to_hex(repl
), path
);
2396 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2397 die("packed object %s (stored in %s) is corrupt",
2398 sha1_to_hex(repl
), p
->pack_name
);
2403 void *read_object_with_reference(const unsigned char *sha1
,
2404 const char *required_type_name
,
2405 unsigned long *size
,
2406 unsigned char *actual_sha1_return
)
2408 enum object_type type
, required_type
;
2410 unsigned long isize
;
2411 unsigned char actual_sha1
[20];
2413 required_type
= type_from_string(required_type_name
);
2414 hashcpy(actual_sha1
, sha1
);
2416 int ref_length
= -1;
2417 const char *ref_type
= NULL
;
2419 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2422 if (type
== required_type
) {
2424 if (actual_sha1_return
)
2425 hashcpy(actual_sha1_return
, actual_sha1
);
2428 /* Handle references */
2429 else if (type
== OBJ_COMMIT
)
2431 else if (type
== OBJ_TAG
)
2432 ref_type
= "object ";
2437 ref_length
= strlen(ref_type
);
2439 if (ref_length
+ 40 > isize
||
2440 memcmp(buffer
, ref_type
, ref_length
) ||
2441 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2446 /* Now we have the ID of the referred-to object in
2447 * actual_sha1. Check again. */
2451 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2452 const char *type
, unsigned char *sha1
,
2453 char *hdr
, int *hdrlen
)
2457 /* Generate the header */
2458 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2462 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2463 git_SHA1_Update(&c
, buf
, len
);
2464 git_SHA1_Final(sha1
, &c
);
2468 * Move the just written object into its final resting place.
2469 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2470 * "moving" is only a part of what it does, when no patch between
2471 * master to pu changes the call sites of this function.
2473 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2477 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2479 else if (link(tmpfile
, filename
))
2483 * Coda hack - coda doesn't like cross-directory links,
2484 * so we fall back to a rename, which will mean that it
2485 * won't be able to check collisions, but that's not a
2488 * The same holds for FAT formatted media.
2490 * When this succeeds, we just return. We have nothing
2493 if (ret
&& ret
!= EEXIST
) {
2495 if (!rename(tmpfile
, filename
))
2499 unlink_or_warn(tmpfile
);
2501 if (ret
!= EEXIST
) {
2502 return error("unable to write sha1 filename %s: %s", filename
, strerror(ret
));
2504 /* FIXME!!! Collision check here ? */
2508 if (adjust_shared_perm(filename
))
2509 return error("unable to set permission to '%s'", filename
);
2513 static int write_buffer(int fd
, const void *buf
, size_t len
)
2515 if (write_in_full(fd
, buf
, len
) < 0)
2516 return error("file write error (%s)", strerror(errno
));
2520 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2521 unsigned char *sha1
)
2525 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2529 /* Finalize a file on disk, and close it. */
2530 static void close_sha1_file(int fd
)
2532 if (fsync_object_files
)
2533 fsync_or_die(fd
, "sha1 file");
2535 die_errno("error when closing sha1 file");
2538 /* Size of directory component, including the ending '/' */
2539 static inline int directory_size(const char *filename
)
2541 const char *s
= strrchr(filename
, '/');
2544 return s
- filename
+ 1;
2548 * This creates a temporary file in the same directory as the final
2551 * We want to avoid cross-directory filename renames, because those
2552 * can have problems on various filesystems (FAT, NFS, Coda).
2554 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2556 int fd
, dirlen
= directory_size(filename
);
2558 if (dirlen
+ 20 > bufsiz
) {
2559 errno
= ENAMETOOLONG
;
2562 memcpy(buffer
, filename
, dirlen
);
2563 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2564 fd
= git_mkstemp_mode(buffer
, 0444);
2565 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2566 /* Make sure the directory exists */
2567 memcpy(buffer
, filename
, dirlen
);
2568 buffer
[dirlen
-1] = 0;
2569 if (mkdir(buffer
, 0777) || adjust_shared_perm(buffer
))
2573 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2574 fd
= git_mkstemp_mode(buffer
, 0444);
2579 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2580 const void *buf
, unsigned long len
, time_t mtime
)
2583 unsigned char compressed
[4096];
2586 unsigned char parano_sha1
[20];
2588 static char tmp_file
[PATH_MAX
];
2590 filename
= sha1_file_name(sha1
);
2591 fd
= create_tmpfile(tmp_file
, sizeof(tmp_file
), filename
);
2593 if (errno
== EACCES
)
2594 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2596 return error("unable to create temporary file: %s", strerror(errno
));
2600 memset(&stream
, 0, sizeof(stream
));
2601 git_deflate_init(&stream
, zlib_compression_level
);
2602 stream
.next_out
= compressed
;
2603 stream
.avail_out
= sizeof(compressed
);
2606 /* First header.. */
2607 stream
.next_in
= (unsigned char *)hdr
;
2608 stream
.avail_in
= hdrlen
;
2609 while (git_deflate(&stream
, 0) == Z_OK
)
2611 git_SHA1_Update(&c
, hdr
, hdrlen
);
2613 /* Then the data itself.. */
2614 stream
.next_in
= (void *)buf
;
2615 stream
.avail_in
= len
;
2617 unsigned char *in0
= stream
.next_in
;
2618 ret
= git_deflate(&stream
, Z_FINISH
);
2619 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2620 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2621 die("unable to write sha1 file");
2622 stream
.next_out
= compressed
;
2623 stream
.avail_out
= sizeof(compressed
);
2624 } while (ret
== Z_OK
);
2626 if (ret
!= Z_STREAM_END
)
2627 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2628 ret
= git_deflate_end_gently(&stream
);
2630 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2631 git_SHA1_Final(parano_sha1
, &c
);
2632 if (hashcmp(sha1
, parano_sha1
) != 0)
2633 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
2635 close_sha1_file(fd
);
2640 utb
.modtime
= mtime
;
2641 if (utime(tmp_file
, &utb
) < 0)
2642 warning("failed utime() on %s: %s",
2643 tmp_file
, strerror(errno
));
2646 return move_temp_to_file(tmp_file
, filename
);
2649 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2651 unsigned char sha1
[20];
2655 /* Normally if we have it in the pack then we do not bother writing
2656 * it out into .git/objects/??/?{38} file.
2658 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2660 hashcpy(returnsha1
, sha1
);
2661 if (has_sha1_file(sha1
))
2663 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2666 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2670 enum object_type type
;
2675 if (has_loose_object(sha1
))
2677 buf
= read_packed_sha1(sha1
, &type
, &len
);
2679 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2680 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2681 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2687 int has_pack_index(const unsigned char *sha1
)
2690 if (stat(sha1_pack_index_name(sha1
), &st
))
2695 int has_sha1_pack(const unsigned char *sha1
)
2697 struct pack_entry e
;
2698 return find_pack_entry(sha1
, &e
);
2701 int has_sha1_file(const unsigned char *sha1
)
2703 struct pack_entry e
;
2705 if (find_pack_entry(sha1
, &e
))
2707 return has_loose_object(sha1
);
2710 static void check_tree(const void *buf
, size_t size
)
2712 struct tree_desc desc
;
2713 struct name_entry entry
;
2715 init_tree_desc(&desc
, buf
, size
);
2716 while (tree_entry(&desc
, &entry
))
2718 * tree_entry() will die() on malformed entries */
2722 static void check_commit(const void *buf
, size_t size
)
2725 memset(&c
, 0, sizeof(c
));
2726 if (parse_commit_buffer(&c
, buf
, size
))
2727 die("corrupt commit");
2730 static void check_tag(const void *buf
, size_t size
)
2733 memset(&t
, 0, sizeof(t
));
2734 if (parse_tag_buffer(&t
, buf
, size
))
2738 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
2739 enum object_type type
,
2740 const char *path
, unsigned flags
)
2742 int ret
, re_allocated
= 0;
2743 int write_object
= flags
& HASH_WRITE_OBJECT
;
2749 * Convert blobs to git internal format
2751 if ((type
== OBJ_BLOB
) && path
) {
2752 struct strbuf nbuf
= STRBUF_INIT
;
2753 if (convert_to_git(path
, buf
, size
, &nbuf
,
2754 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
2755 buf
= strbuf_detach(&nbuf
, &size
);
2759 if (flags
& HASH_FORMAT_CHECK
) {
2760 if (type
== OBJ_TREE
)
2761 check_tree(buf
, size
);
2762 if (type
== OBJ_COMMIT
)
2763 check_commit(buf
, size
);
2764 if (type
== OBJ_TAG
)
2765 check_tag(buf
, size
);
2769 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2771 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2777 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
2778 const char *path
, unsigned flags
)
2780 struct strbuf sbuf
= STRBUF_INIT
;
2783 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2784 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
2787 strbuf_release(&sbuf
);
2791 #define SMALL_FILE_SIZE (32*1024)
2793 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
2794 enum object_type type
, const char *path
,
2800 ret
= index_mem(sha1
, NULL
, size
, type
, path
, flags
);
2801 } else if (size
<= SMALL_FILE_SIZE
) {
2802 char *buf
= xmalloc(size
);
2803 if (size
== read_in_full(fd
, buf
, size
))
2804 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
2806 ret
= error("short read %s", strerror(errno
));
2809 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2810 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
2817 * This creates one packfile per large blob unless bulk-checkin
2818 * machinery is "plugged".
2820 * This also bypasses the usual "convert-to-git" dance, and that is on
2821 * purpose. We could write a streaming version of the converting
2822 * functions and insert that before feeding the data to fast-import
2823 * (or equivalent in-core API described above). However, that is
2824 * somewhat complicated, as we do not know the size of the filter
2825 * result, which we need to know beforehand when writing a git object.
2826 * Since the primary motivation for trying to stream from the working
2827 * tree file and to avoid mmaping it in core is to deal with large
2828 * binary blobs, they generally do not want to get any conversion, and
2829 * callers should avoid this code path when filters are requested.
2831 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
2832 enum object_type type
, const char *path
,
2835 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
2838 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
2839 enum object_type type
, const char *path
, unsigned flags
)
2842 size_t size
= xsize_t(st
->st_size
);
2844 if (!S_ISREG(st
->st_mode
))
2845 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
2846 else if (size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
2847 (path
&& would_convert_to_git(path
, NULL
, 0, 0)))
2848 ret
= index_core(sha1
, fd
, size
, type
, path
, flags
);
2850 ret
= index_stream(sha1
, fd
, size
, type
, path
, flags
);
2855 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
2858 struct strbuf sb
= STRBUF_INIT
;
2860 switch (st
->st_mode
& S_IFMT
) {
2862 fd
= open(path
, O_RDONLY
);
2864 return error("open(\"%s\"): %s", path
,
2866 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
2867 return error("%s: failed to insert into database",
2871 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
2872 char *errstr
= strerror(errno
);
2873 return error("readlink(\"%s\"): %s", path
,
2876 if (!(flags
& HASH_WRITE_OBJECT
))
2877 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
2878 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
2879 return error("%s: failed to insert into database",
2881 strbuf_release(&sb
);
2884 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2886 return error("%s: unsupported file type", path
);
2891 int read_pack_header(int fd
, struct pack_header
*header
)
2893 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2894 /* "eof before pack header was fully read" */
2895 return PH_ERROR_EOF
;
2897 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2898 /* "protocol error (pack signature mismatch detected)" */
2899 return PH_ERROR_PACK_SIGNATURE
;
2900 if (!pack_version_ok(header
->hdr_version
))
2901 /* "protocol error (pack version unsupported)" */
2902 return PH_ERROR_PROTOCOL
;
2906 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
2908 enum object_type type
= sha1_object_info(sha1
, NULL
);
2910 die("%s is not a valid object", sha1_to_hex(sha1
));
2912 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),